Design And Performance Optimization Of Internet Advertising Server Cluster

With the rapid development of the Internet,the number of domestic Internet users has grown rapidly,many advertisers have also shifted their advertising from traditional channels such as TV broadcasting to Internet.Since the 21st century,thanks to the rapid development of computing advertising technology and the high profit margin of its own business,major domestic e-commerce giants,mainstream search engines and large-scale portals have set up their own advertising platform.For example,Baidu DSP of Baidu,Ali Mama of Alibaba Group,etc.,more than ten years of practice also prove that the most effective business model of the Internet is the online advertising mode that can directly realize the traffic.In the ten years of this big wave,the Internet advertising system has become more and more stable,forming a system with Ad Exchange as the core hub,DSP(demand side platform)and SSP(supply side platform)as the intermediate nodes,and associated advertisers and media delivery Structure.With the rise and development of the concept of programmatic purchase,the entire advertising process puts higher demands on DSP.An excellent DSP platform not only has strong RTB(Real-Time Bidding)capability,but also needs accurate crowd orientation technology.This means that the DSP platform must face both high-concurrency and super-large computational coexistence,which not only challenges the server performance,but also leaves a problem for server and algorithm developersSince I was a server architect on the demand side platform of a domestic first-line Internet company during my postgraduate studies,this article will focus on the DSP platform as the core,including the platform architecture,transaction methods and data storage.Through the analysis of core performance indicators such as precise orientation,real-time bidding and other core services,server load,request duration,etc.,and integrating my work experience,I first explain the reasons for the passive loss of AOAS server clusters in this system.Combining the server structure with the executed business logic,the caching strategy based on Redis in-memory database and Kafka message middleware is given and finally achieved good results;and according to the feedback from colleagues of user data department that "some products and users are rarely involved in the crowd orientation",the analysis report and solution are given for this situation.Since the author is not responsible for this business and does not involve the orientation and the material library module,it is finally discussed and decided to start the solutionfrom the ARAS server in the implementation framework of this paper,combined with the existing "The problemof lacking of orientation accuracyin dCPD advertising",the solution of directional bottoming is given.The final experimental results prove that this scheme works well without modifying the orientation module and the degree has alleviated the above problem and improved the orientation accuracy of dCPD advertising.